[Python-Dev] Test failures under Windows?

[David Bolen]

Curt Hagenlocher <curt at hagenlocher.org> writes:

> The variation that goes through assert.c should write to stderr for a
> console-mode application, so it's reasonable to assume that we're
> hitting the other code path -- and that Mark's suggestion to use
> CrtSetReportMode would address the issue.

Which is pretty much what the regrtest.py -n option does/did in the
py3k branch (and release30-maint).  It enables the following code:

        import msvcrt
        msvcrt.SetErrorMode(msvcrt.SEM_FAILCRITICALERRORS|
                msvcrt.SEM_NOALIGNMENTFAULTEXCEPT|
                msvcrt.SEM_NOGPFAULTERRORBOX|
                msvcrt.SEM_NOOPENFILEERRORBOX)
        try:
            msvcrt.CrtSetReportMode
        except AttributeError:
            # release build
            pass
        else:
            for m in [msvcrt.CRT_WARN, msvcrt.CRT_ERROR, msvcrt.CRT_ASSERT]:
                msvcrt.CrtSetReportMode(m, msvcrt.CRTDBG_MODE_FILE)
                msvcrt.CrtSetReportFile(m, msvcrt.CRTDBG_FILE_STDERR)

which seems like it would still be the right sort of code to use.  For
my buildbot the SetErrorMode is redundant because I already do that in the
buildbot code itself (and the setting is inherited) but it won't hurt.

>From what I can see though, the tools/buildbot/test.bat file no longer
adds the -n option that it used to, although I'm unclear on why it
might have been removed.  Perhaps this was just a regression that was
accidentally missed, as it appears to have disappeared during a large
merger from trunk to the py3k branch mid-2008 (r64273) when the batch
file line ending was changed to CRLF.

It would be nice to also have this in the other active release branches.

-- David